Contexts update + buffer flushing + cleanup (#338)

* Update lib/rocprofiler-sdk/context/context.*

- get_registered_contexts functions (local copy)

* Update lib/rocprofiler-sdk/hsa/{queue,queue_controller}.cpp

- remove ROCPROFILER_BUFFER_TRACING_MEMORY_COPY code

* Update tests/kernel-tracing/kernel-tracing.cpp

- move stop() and flush() in tool_fini to before reporting of sizes of data collected

* Update lib/rocprofiler-sdk/hsa/hsa.*

- remove stale set_callback / activity_functor_t code

* Update lib/rocprofiler-sdk/buffer.cpp

- full wait instead of returning busy when buffer is busy
- use task_group::join instead of task_group::wait to fully wait for tasks to finish (bug fix)

* Update lib/rocprofiler-sdk/agent.cpp

- support agent mapping for CPU agents

* Remove direct access to vector of registered contexts

tests/kernel-tracing/kernel-tracing.cpp

@@ -642,6 +642,9 @@ tool_fini(void* tool_data)
     static auto _once = std::atomic_flag{ATOMIC_FLAG_INIT};
     if(_once.test_and_set()) return;
 
+    stop();
+    flush();
+
     std::cerr << "[" << getpid() << "][" << __FUNCTION__
               << "] Finalizing... agents=" << agents.size()
               << ", code_object_callback_records=" << code_object_records.size()
@@ -652,9 +655,6 @@ tool_fini(void* tool_data)
               << ", hsa_api_bf_records=" << hsa_api_bf_records.size() << " ...\n"
               << std::flush;
 
-    stop();
-    flush();
-
     auto* _call_stack = static_cast<call_stack_t*>(tool_data);
     if(_call_stack)
     {
@@ -799,6 +799,7 @@ flush()
 {
     for(auto* itr : buffers)
     {
+        if(!itr) continue;
         auto status = rocprofiler_flush_buffer(*itr);
         if(status != ROCPROFILER_STATUS_ERROR_BUFFER_BUSY)
         {